Remote health monitoring – a systems approach to using IoT Technologies

Poyner, I. K. (2023) Remote health monitoring – a systems approach to using IoT Technologies. PhD thesis, University of Reading. doi: 10.48683/1926.00119533

Abstract/Summary

Remote Health Monitoring (RHM) has benefitted greatly from powerful smartphones and the very high data rate mobile networks. However, RHM benefits may be unobtainable for many users living with health challenges or in remote areas not served by telecommunications companies. IoT (Internet of Things) systems may redress some of these inequalities and extend RHM to a much wider community of users. This thesis takes a systems-engineering approach to consider the service as a whole, and identifies the regulatory, business and user needs, especially reliability and privacy of personal health information. Relevant frameworks and technical requirements are assessed for a constrained device, including energy efficiency and security. IoT networks, such as LoRaWAN, provide options for low cost, low power data transfer which are secure and do not depend upon network operators, especially when transmitted via satellites. Additionally, Machine Learning (ML) on constrained embedded devices is now practical, further reducing the need to transmit data for off-board processing. However, challenges remain for providing reliable and adaptable services to users whose health, and potentially life, relies on RHM services. Regulators are providing guidelines, but it is probable that legislation may in future enforce this guidance. A TI CC2652 board was used to practically measure the relative energy consumption of transmitting packets of data via Bluetooth Low Energy (BLE) compared to on-board processing. A BLE message with a data payload of MTU = 251 bytes consumes approximately 660 – 676 nJ, which will also be dependent upon transmitted signal strength. This equates approximately to the CPU processing 11,380 – 11,655 for-loops. This provides a metric by which specific on-board processing and machine learning strategies can be assessed as to their energy efficiencies compared to offloading the raw data for processing. Advancements in ML for edge devices, such as TinyML and TensorFlow Lite for Microcontrollers, may enable very specific models to be run on the device within this energy budget. For comparison, this is approximately 50 times lower than the energy consumption of a BLE triple advertisement by the SPHERE SPW-1 wearable which consumes between 37 µJ (at -20dBm) and 60 µJ (at 4 dBm).

Item Type	Thesis (PhD)
URI	https://reading-clone.eprints-hosting.org/id/eprint/119533
Identification Number/DOI	10.48683/1926.00119533
Divisions	Life Sciences > School of Biological Sciences > Department of Bio-Engineering
Download/View statistics	View download statistics for this item